Generalized Multiscale Finite Element Methods. Nonlinear ...
MULTISCALE COMPUTATIONAL METHODS
description
Transcript of MULTISCALE COMPUTATIONAL METHODS
MULTISCALE COMPUTATIONAL
METHODS
Achi BrandtThe Weizmann Institute of ScienceUCLA
www.wisdom.weizmann.ac.il/~achi
0u 1u2u
3u4u
given
02
043214
Fh
uuuuu
Poisson equation:2 2
2 2
u uF
x y
Approximating Poisson equation:
given
Solving PDE: Influence of pointwiserelaxation on the error
Error of initial guess Error after 5 relaxation sweeps
Error after 10 relaxations Error after 15 relaxations
Fast error smoothingslow solution
When relaxation slows down:
the error is a sum of low eigen-vectors
ELLIPTIC PDE'S (e.g., Poisson equation)
the error is smooth
The error can be approximated on a coarser grid
LU=F
h
2h
4h
LhUh=Fh
L2hU2h=F2h
L4hU4h=F4h
hLhUh = Fh
LocalRelaxation
hu~approximation
hV hh u~U smooth
2h
4h
L2hU2h = F2h
L4hV4h = R4h
L2hV2h = R2h R2h = ( Fh -Lh )hh2 hu~
interpolation (order l+p)to a new grid
interpolation (order m)of corrections
relaxation sweeps
algebraic error< truncation error
residual transfer
ν νenough sweepsor direct solver*
2ν
Full MultiGrid (FMG) algorithm
..
.
*
1ν
1ν
1ν
2ν
2ν
2ν
h0
h0/2
h0/4
2h
h
**
1ν
1ν
1ν2ν
*
2ν
2ν
2ν
multigrid cycle V
Multigrid solversCost: 25-100 operations per unknown
• Linear scalar elliptic equation (~1971)
Multigrid solversCost: 25-100 operations per unknown
• Linear scalar elliptic equation (~1971)*• Nonlinear• Grid adaptation• General boundaries, BCs*• Discontinuous coefficients• Disordered: coefficients, grid (FE) AMG• Several coupled PDEs* (1980)
• Non-elliptic: high-Reynolds flow• Highly indefinite: waves• Many eigenfunctions (N)• Near zero modes• Gauge topology: Dirac eq.• Inverse problems• Optimal design• Integral equations Full matrix• Statistical mechanics
Massive parallel processing*Rigorous quantitative analysis
(1986)
FAS (1975)
Within one solver
)log(
2
NNO
fuku
(1977,1982)
Multigrid solversCost: 25-100 operations per unknown
• Linear scalar elliptic equation (~1971)*• Nonlinear• Grid adaptation• General boundaries, BCs*• Discontinuous coefficients• Disordered: coefficients, grid (FE) AMG• Several coupled PDEs* (1980)
• Non-elliptic: high-Reynolds flow• Highly indefinite: waves• Many eigenfunctions (N)• Near zero modes• Gauge topology: Dirac eq.• Inverse problems• Optimal design• Integral equations• Statistical mechanics
Massive parallel processing*Rigorous quantitative analysis
(1986)
FAS (1975)
Within one solver
)log(
2
NNO
fuku
(1977,1982)
interpolation (order l+p)to a new grid
interpolation (order m)of corrections
relaxation sweeps
algebraic error< truncation error
residual transfer
ν νenough sweepsor direct solver*
**
1ν
1ν
1ν2ν
*
2ν
2ν
2ν
Full MultiGrid (FMG) algorithm
..
.
*
1ν
1ν
1ν
2ν
2ν
2ν
Vcyclemultigrid
h0
h0/2
h0/4
2h
h
h
2h
4h
LhUh = Fh
L4hU4h = F4h
h2
h4
Fine-to-coarse defect correction
L2hV2h = R2h
Full Approximation scheme (FAS):
U2h = + V2h L2hU2h = F2h
LocalRelaxation
hu~approximation
hV hh u~U smooth
hu~hh2
Multigrid solversCost: 25-100 operations per unknown
• Linear scalar elliptic equation (~1971)*• Nonlinear• Grid adaptation• General boundaries, BCs*• Discontinuous coefficients• Disordered: coefficients, grid (FE)• Several coupled PDEs*• Non-elliptic: high-Reynolds flow• Highly indefinite: waves• Many eigenfunctions (N)• Near zero modes• Gauge topology: Dirac eq.• Inverse problems• Optimal design• Integral equations• Statistical mechanics
Massive parallel processing*Rigorous quantitative analysis
Within one solver
interpolation (order l+p)to a new grid
interpolation (order m)of corrections
relaxation sweeps
algebraic error< truncation error
residual transfer
ν νenough sweepsor direct solver*
**
1ν
1ν
1ν2ν
*
2ν
2ν
2ν
Full MultiGrid (FMG) algorithm
..
.
*
1ν
1ν
1ν
2ν
2ν
2ν
Vcyclemultigrid
h0
h0/2
h0/4
2h
h
h
2h
4h
LhUh = Fh
L4hU4h = F4h
h2
h4
Fine-to-coarse defect correction
L2hV2h = R2h
Full Approximation scheme (FAS):
U2h = + V2h L2hU2h = F2h
LocalRelaxation
hu~approximation
hV hh u~U smooth
Truncation error estimator
hu~hh2
Multigrid solversCost: 25-100 operations per unknown
• Linear scalar elliptic equation (~1971)*• Nonlinear• Grid adaptation• General boundaries, BCs*• Discontinuous coefficients• Disordered: coefficients, grid (FE) AMG• Several coupled PDEs*
(1980)
• Non-elliptic: high-Reynolds flow• Highly indefinite: waves• Many eigenfunctions (N)• Near zero modes• Gauge topology: Dirac eq.• Inverse problems• Optimal design• Integral equations• Statistical mechanics
Massive parallel processing*Rigorous quantitative analysis
(1986)
FAS (1975)
Within one solver
)log(
2
NNO
fuku
(1977,1982)
• Same fast solver FMG
Local patches of finer grids
• Each patch may use different coordinate system and anisotropic grid and different
physics; e.g. Atomistic
interpolation (order l+p)to a new grid
interpolation (order m)of corrections
relaxation sweeps
algebraic error< truncation error
residual transfer
ν νenough sweepsor direct solver*
2ν
Full MultiGrid (FMG) algorithm
..
.
*
1ν
1ν
1ν
2ν
2ν
2ν
h0
h0/2
h0/4
2h
h
**
1ν
1ν
1ν2ν
*
2ν
2ν
2ν
multigrid cycle V
• Same fast solver FMG,
Local patches of finer grids
• Each level correct the equations of the next coarser level
• Each patch may use different coordinate system and anisotropic grid
• Same fast solver FMG, FAS
x , y
)(
)(
sy
sx
r , s
Finer level with local coordinate transformation
Boundary or tracked layer
With anisotropic further refinement
• Same fast solver FMG,
Local patches of finer grids
• Each level correct the equations of the next coarser level
• Each patch may use different coordinate system and anisotropic grid
“Quasicontiuum” method [B., 1992]
• Each patch may use different coordinate system and anisotropic grid and different
physics; e.g. Atomistic
and differet physics; e.g. atomistic
• Same fast solver FMG, FAS
Multigrid solversCost: 25-100 operations per unknown
• Linear scalar elliptic equation (~1971)*• Nonlinear• Grid adaptation• General boundaries, BCs*• Discontinuous coefficients• Disordered: coefficients, grid (FE)• Several coupled PDEs* (1980)• Non-elliptic: high-Reynolds flow• Highly indefinite: waves• Many eigenfunctions (N)• Near zero modes• Gauge topology: Dirac eq.• Inverse problems• Optimal design• Integral equations• Statistical mechanics
Massive parallel processing*Rigorous quantitative analysis
(1986)
Stokes
0
0
0
yx
y
x
vu
Pv
Pu
0
0
0
0
0
0
p
v
u
yx
y
x
2 yyxxL
L
det
h-principal LLu f Ldet
0
v
u
Riemann
Cauchy
xy
yx
Compressible Navier-Stokes(on the viscous scale)
uk3
2
Central Cauchy-Riemannh2
Central (Navier-) Stokeshh
Q2
Stokes
2D
0
0
0
yx
y
x
20
p
v
u
StokesNavier
ibleIncompress
2D
Q
0
0
0
yx
y
x
Q
Q
1Q u R
p
v
u
Euler
2D
1
0
00
0
0
0
0
0
0
0
0
0 y
x
y
y
x
x
P
u
P
u
u
u
22
23
au
xu
PPa p2
2
p
v
u
Multigrid solversCost: 25-100 operations per unknown
• Linear scalar elliptic equation (~1971)*• Nonlinear• Grid adaptation• General boundaries, BCs*• Discontinuous coefficients• Disordered: coefficients, grid (FE) • Several coupled PDEs* • Non-elliptic: high-Reynolds flow• Highly indefinite: waves• Many eigenfunctions (N)• Near zero modes• Gauge topology: Dirac eq.• Inverse problems• Optimal design• Integral equations• Statistical mechanics
Massive parallel processing*Rigorous quantitative analysis
Multigrid solversCost: 25-100 operations per unknown
• Linear scalar elliptic equation (~1971)*• Nonlinear• Grid adaptation• General boundaries, BCs*• Discontinuous coefficients• Disordered: coefficients, grid (FE)• Several coupled PDEs*• Non-elliptic: high-Reynolds flow• Highly indefinite: waves• Many eigenfunctions (N)• Near zero modes• Gauge topology: Dirac eq.• Inverse problems• Optimal design• Integral equations• Statistical mechanics
Massive parallel processing*Rigorous quantitative analysis
ALGEBRAIC MULTIGRID (AMG) 1982
Ax = b
When relaxation slows down:
DISCRETIZED PDE'S
GENERAL SYSTEMS OF LOCAL EQUATIONS
the error is smooth
Along characteristics
The error can be approximated
by a far fewer degrees
of freedom (coarser grid)
the error is a sum of low eigen-vectors
ELLIPTIC PDE'S
the error is smooth
ALGEBRAIC MULTIGRID (AMG) 1982
Coarse variables - a subset
Criterion: Fast convergence of “compatible relaxation”
Ax = b
Relax Ax = 0Keeping coarse variables = 0
ALGEBRAIC MULTIGRID (AMG) 1982
Coarse variables - a subset
1. “General” linear systems2. Variety of graph problems
General procedures for deriving:* Interpolations
Ax = b
* Restriction* Coarse-level equations
Generalizations:
Graph problems
Partition: min cut
Clustering bioinformatics
Image segmentation
VLSI placement Routing
Linear arrangement: bandwidth, cutwidth
Graph drawing low dimension embedding